Yield prediction and optimization of biomass-based products by multi-machine learning schemes: Neural, regression and function-based techniques

Pyrolysis, as a thermochemical conversion of biomass, is a superior biofuel production procedure. The deter-mining procedure for the optimal operational parameters, biomass characteristics, and types is outstandingly complex. Machine learning (ML) models were applied to enhance the predictive performance of three biofuel yields (bio-char, bio-oil, and syngas). This study aimed to establish seven ML models by utilizing the extracted experimental datasets of various pyrolysis routes of biomass (walnut shells and seed cake). The yields of three biofuels are mostly estimated at 0.95 to 0.99 of R-squared. Moreover, the sensitivity analysis displayed that species of biomass and pyrolysis conditions exhibited high errors (5.26-5.62% and 2-4.63% of MAPE) by excluding them from the input set for yield predictions. The generalizability of the ML technique is observed. The radial basis function (RBF) is highly capable of estimating biofuel yield. Genetic algorithms based on radial basis function (GA-RBF) optimization are applied in two ways: single bio-fuel and biomass species. The optimal yields achieved 36.04, 45, and 54.16% for the three biofuels, respectively. Three types of ML demonstrated the high feasibility of biofuel yield prediction. The findings provide strong evidence for using the potential of ML as an assistant along with desirable biofuel production.

Automated summary

This study utilized machine learning models to enhance the predictive performance of biofuel yields in biomass pyrolysis. The results showed that the radial basis function (RBF) performed well in estimating biofuel yield. The findings suggest that machine learning has great potential as an assistant in biofuel production.